A method, medium and system for predicting cracks in civil construction wall

Abstract

The present invention provides a method, medium and system for predicting cracks in a wall during civil construction, which belongs to the technical field of civil construction, and includes: firstly, based on known wall structure information, a three-dimensional finite element model is established, and stress concentration, structural defects and crack-prone areas are determined; next, multi-mode images of the wall are acquired regularly, and these images are pre-processed and feature fused to obtain a time series feature matrix; for these feature matrices, a difference matrix sequence between adjacent time points is calculated, and the matrix with the largest absolute value is selected as the maximum change matrix. Combined with the information of the crack-prone area, the maximum change matrix is ​​screened to obtain a focus matrix. The focus matrix is ​​input into a pre-trained crack prediction model, and the crack risk score of each crack-prone area is output. Finally, the crack-prone areas are graded according to the crack risk score, and the areas with higher risks are listed as key monitoring areas, thereby realizing accurate prediction and monitoring of wall crack risks.

Description

Technical Field

[0001] The present invention belongs to the technical field of civil construction, and in particular, relates to a method, medium and system for predicting cracks in a wall during civil construction. Background Art

[0002] The problem of wall cracks has always been a focus of attention in the field of construction and maintenance. Wall cracks not only affect the aesthetics of the building, but may also threaten the overall structural safety of the building. Therefore, timely detection and prevention of wall cracks is crucial to ensure the service life and safety of the building.

[0003] At present, common wall crack detection methods mainly include manual inspection and single sensor monitoring. Manual inspection relies on the experience and visual inspection of professional inspectors, with low detection efficiency and certain subjectivity; while single sensor monitoring, such as strain sensors or infrared thermal imagers, can obtain certain crack information in real time, but due to the single monitoring indicator, it is difficult to fully reflect the occurrence and development of cracks.

[0004] In addition, even if wall cracks are found, how to accurately assess their occurrence risk and evolution trend is also an urgent problem to be solved. Existing crack assessment methods usually rely on empirical judgment or simple threshold analysis, which makes it difficult to objectively reflect the actual degree of damage caused by cracks. Summary of the invention

[0005] In view of this, the present invention provides a method, medium and system for predicting cracks in civil construction walls, which can solve the technical problem that existing crack assessment methods usually rely on empirical judgment or simple threshold analysis and are difficult to objectively reflect the actual degree of damage of cracks.

[0006] The present invention is achieved in that:

[0007] A first aspect of the present invention provides a method for predicting cracks in a wall during civil construction, comprising the following steps:

[0008] S10, establishing a three-dimensional finite element model of the wall according to known wall structure information;

[0009] S20, according to the three-dimensional finite element model of the wall, determining areas of stress concentration, structural defects, or cracked areas in the wall as key crack-prone areas;

[0010] S30, acquiring and preprocessing multi-mode images of the wall at specified time intervals, including ultrasonic reflection images, vibration transmission images, and surface infrared images;

[0011] S40, extracting and fusing the preprocessed ultrasonic reflection image, vibration transmission image, and surface infrared image to obtain a fusion feature matrix;

[0012] S50, for multiple fusion feature matrices obtained in time series, calculate the difference between matrices at adjacent time points to obtain a difference matrix sequence, and select a difference matrix with the largest matrix norm from the difference matrix sequence as a maximum change matrix;

[0013] S60, screening the maximum change matrix according to the crack-prone area to obtain a focus matrix;

[0014] S70, inputting the attention matrix into a pre-trained crack prediction model to obtain a crack risk matrix for representing the crack risk score of each crack-prone area;

[0015] S80. Crack-prone areas are classified according to the crack risk scores, and crack-prone areas with high scores are listed as key monitoring areas and output.

[0016] The step S10 specifically includes:

[0017] Step 1: discretize the three-dimensional geometric configuration of the wall into three-dimensional unit grids, such as tetrahedral units or hexahedral units, according to known wall material properties, geometric dimensions and support conditions;

[0018] Step 2: Assign corresponding mechanical parameters to each unit, such as elastic modulus, Poisson's ratio, etc., to reflect the characteristics of the wall material;

[0019] Step 3: Determine the load and boundary conditions of the finite element model based on the support conditions of the wall, such as boundary constraints;

[0020] Step 4: Use the finite element solution method to calculate the stress distribution, deformation field and other physical quantities of the wall under various loads, and complete the establishment of the three-dimensional finite element model.

[0021] Wherein, the step S20 specifically includes:

[0022] Step 201, extracting the stress tensor of each unit from the three-dimensional finite element model of the wall established in step S10;

[0023] Step 202, calculating the equivalent stress of each unit;

[0024] Step 203: According to a preset stress concentration threshold, the unit whose equivalent stress is greater than the stress concentration threshold is divided into a stress concentration area;

[0025] Step 204, based on known structural defect features, including at least holes and defect depth, divide the unit containing such defects into structural defect areas;

[0026] Step 205: Combine the stress concentration area and the structural defect area to determine the crack-prone area in the wall.

[0027] The equivalent stress adopts the von Mises equivalent stress.

[0028] Wherein, the step S30 specifically includes:

[0029] Step 301, using a pulse echo method to obtain an ultrasonic reflection image of the internal structure of the wall;

[0030] Step 302: Use a time window function to perform windowing processing on the ultrasonic echo signal to filter out irrelevant interference signals;

[0031] Step 303: Use a contact acceleration sensor to measure the vibration response of the wall surface, and use Fourier transform to convert the time domain signal into frequency domain features to obtain a vibration transmission image;

[0032] Step 304: Use an infrared thermal imager to capture the temperature distribution on the wall surface, and perform radiation correction and pseudo-color processing to obtain a surface infrared image;

[0033] Step 305: Repeat steps 1-4 at specified time intervals to acquire and preprocess a set of multi-mode image data.

[0034] The specified time interval is generally 1 hour to 6 hours.

[0035] Wherein, the step S40 specifically includes:

[0036] Step 401, extracting background noise, attenuation, and echo peak features from the ultrasonic reflection image;

[0037] Step 402: extracting peak frequency and power spectrum density characteristics of the vibration spectrum from the vibration transmission image;

[0038] Step 403: extracting the mean, variance, and gradient features of the temperature distribution from the infrared thermal imaging image;

[0039] Step 404: Use principal component analysis or independent component analysis to integrate the above multi-modal image features into a fusion feature matrix.

[0040] Wherein, the step S50 specifically includes:

[0041] Step 501: Let the fusion feature matrix at a certain time t be X(t), and calculate the difference matrix ΔX(t,t+Δt)=X(t+Δt)-X(t) between adjacent times t and t+Δt;

[0042] Step 502, calculate the difference matrix of all adjacent moments to form a difference matrix sequence {ΔX(t,t+Δt)};

[0043] Step 503: Select the difference matrix ΔX with the largest matrix norm from the difference matrix sequence max Maximum change matrix.

[0044] The matrix norm here can adopt the L2 norm.

[0045] Furthermore, the step S60 specifically includes:

[0046] Step 601: The maximum change matrix ΔX obtained in step S50 is max Sort by the absolute value of each element in;

[0047] Step 602, select the row and column coordinates of the first n% of elements, which are the coordinates of the focus area;

[0048] Step 603: Extract the corresponding attention matrix ΔX according to the crack-prone area determined in step S20 focus .

[0049] Usually n can be 10% to 20%.

[0050] Furthermore, the step S70 specifically includes:

[0051] Step 701: The attention matrix ΔX obtained in step S60 focus Input into the pre-trained crack prediction model;

[0052] Step 702: The crack prediction model outputs the crack risk score of each crack-prone area to form a crack risk matrix.

[0053] Furthermore, the step S80 specifically includes:

[0054] Step 801: Set a crack risk threshold R th , the risk score is greater than R th areas are classified as high-risk areas;

[0055] Step 802: sort the high-risk areas according to their risk scores, and select the top m% of areas as key monitoring areas;

[0056] Step 803: Output the information of the key monitoring area.

[0057] Usually m can be taken as 20%; R th =0.7.

[0058] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for predicting cracks in a civil construction wall.

[0059] A third aspect of the present invention provides a civil construction wall crack prediction system, which includes the above-mentioned computer-readable storage medium.

[0060] Compared with the prior art, the method, medium and system for predicting cracks in civil construction walls provided by the present invention have the following beneficial effects:

[0061] 1. Enhanced comprehensive perception capability: This method uses multiple physical quantity information such as ultrasonic reflection, vibration response, infrared thermal imaging, etc. to more comprehensively perceive the structural characteristics and mechanical state of the wall. Compared with single sensor monitoring, this multimodal data fusion can more accurately reflect the actual situation of the wall and provide richer and more accurate basic data for subsequent crack prediction.

[0062] 2. Improved identification and early warning capabilities: Based on the three-dimensional finite element model and feature analysis, this method can accurately identify stress concentration areas and structural defect areas in the wall, which are often prone to cracks. By focusing on monitoring and early warning these key areas, the timeliness and accuracy of crack detection can be greatly improved.

[0063] 3. Significantly improved prediction accuracy: This method uses a crack prediction model based on machine learning, which can more objectively and accurately assess the crack risk of each monitoring area. Compared with empirical judgment or simple threshold analysis, this data-driven prediction method can better capture the complex laws of crack occurrence and development, thereby obtaining more reliable risk assessment results.

[0064] In general, the wall crack prediction method based on multimodal image fusion and machine learning proposed in the present invention makes full use of various types of physical quantity information, realizes comprehensive perception, accurate prediction and automatic monitoring of cracks, greatly improves the efficiency and reliability of crack detection and prevention, and solves the technical problem that existing crack assessment methods usually rely on empirical judgment or simple threshold analysis, which makes it difficult to objectively reflect the actual degree of harm of cracks. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0066] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0068] like Figure 1 FIG. 1 is a flow chart of a method for predicting cracks in a wall during civil construction provided by the present invention. The method comprises the following steps:

[0069] S10, establishing a three-dimensional finite element model of the wall according to known wall structure information;

[0070] S20, according to the three-dimensional finite element model of the wall, determining areas of stress concentration, structural defects, or cracked areas in the wall as key crack-prone areas;

[0071] S30, acquiring and preprocessing multi-mode images of the wall at specified time intervals, including ultrasonic reflection images, vibration transmission images, and surface infrared images;

[0072] S40, extracting and fusing the preprocessed ultrasonic reflection image, vibration transmission image, and surface infrared image to obtain a fusion feature matrix;

[0073] S50, for multiple fusion feature matrices obtained in time series, calculate the difference between matrices at adjacent time points to obtain a difference matrix sequence, and select a difference matrix with the largest matrix norm from the difference matrix sequence as a maximum change matrix;

[0074] S60, screening the maximum change matrix according to the crack-prone area to obtain a focus matrix;

[0075] S70, inputting the attention matrix into a pre-trained crack prediction model to obtain a crack risk matrix for representing the crack risk score of each crack-prone area;

[0076] S80. Crack-prone areas are classified according to the crack risk scores, and crack-prone areas with higher scores are listed as key monitoring areas and output.

[0077] The specific implementation methods of the above steps are described in detail below:

[0078] The specific implementation method of step S10 is: according to the known wall structure information, such as wall material, geometric dimensions, support conditions, etc., a three-dimensional finite element model of the wall is established using the finite element method. Specifically, the geometric configuration of the wall is first discretized and divided into a three-dimensional unit grid, such as a tetrahedron unit or a hexahedron unit; then, according to the characteristics of the wall material, corresponding mechanical parameters are assigned to each unit, such as elastic modulus, Poisson's ratio, etc.; finally, according to the support conditions of the wall, such as boundary constraints, the load and boundary conditions of the model are determined. Through finite element solution, physical quantities such as stress distribution and deformation field of the wall under various loads can be obtained. The purpose of this step is to establish a numerical model that can reflect the real mechanical behavior of the wall, providing a basis for the subsequent identification of stress concentration areas.

[0079] The specific implementation of step S20 is: according to the three-dimensional finite element model of the wall established in step S10, the stress concentration area and the structural defect area in the wall are identified through stress field analysis. Specifically, firstly, the stress tensor of each unit in the model is extracted, and the equivalent stress of each unit is calculated, such as the von Mises equivalent stress; then, according to the pre-set stress concentration threshold σ c , the equivalent stress is greater than σ c The units with structural defects are divided into stress concentration areas; at the same time, according to the known structural defect characteristics, such as holes, defect depth, etc., the units containing such defects are divided into structural defect areas. Through this step, the crack-prone areas with stress concentration and structural defects in the wall can be determined, providing target areas for subsequent crack detection and prediction. In general, σ c It can be about 80% of the tensile strength of the wall material.

[0080] The specific implementation method of step S30 is: within each specified time period (such as 1 hour), obtain and pre-process the multi-mode image data of the wall, including ultrasonic reflection image, vibration transmission image and surface infrared image. Specifically, for ultrasonic reflection image, first use the pulse echo method to obtain the internal structure information of the wall, and then use the time window function to perform windowing processing on the echo signal to filter out irrelevant signals; for vibration transmission image, use a contact acceleration sensor to measure the vibration response of the wall surface, and use Fourier transform to convert the time domain signal into frequency domain features; for infrared thermal imaging image, use an infrared thermal imager to shoot the temperature distribution of the wall surface, and perform radiation correction and pseudo-color processing. The purpose of this step is to obtain multi-mode image data reflecting the internal structure, dynamic characteristics and thermal characteristics of the wall, and provide a basis for subsequent feature extraction and crack prediction.

[0081] The specific implementation method of step S40 is: extract and fuse the pre-processed ultrasonic reflection image, vibration transmission image and surface infrared image to obtain a fused feature matrix. Specifically, for the ultrasonic reflection image, extract the background noise, attenuation, echo peak and other features; for the vibration transmission image, extract the peak frequency of the vibration spectrum, power spectrum density and other features; for the infrared thermal imaging image, extract the mean, variance, gradient and other features of the temperature distribution. Then, use the feature fusion method such as principal component analysis (PCA) or independent component analysis (ICA) to integrate the above multi-mode image features into a fused feature matrix. The purpose of this step is to extract features rich in crack information from the perspective of multiple physical quantities to enhance the accuracy of crack prediction.

[0082] The specific implementation method of step S50 is: for multiple fusion feature matrices obtained in time series, calculate the differences between matrices at adjacent time points to obtain a difference matrix sequence, and select the difference matrix with the largest absolute value as the maximum change matrix. Specifically, let the fusion feature matrix at a certain time t be X(t), then the difference matrix between adjacent times t and t+Δt is: ΔX(t, t+Δt) = X(t+Δt)-X(t). Calculate the difference matrices of all adjacent times to form a difference matrix sequence {ΔX(t, t+Δt)}. Select the difference matrix with the largest matrix norm (such as L2 norm) from the sequence, and record it as ΔX max Maximum change matrix. The purpose of this step is to extract the most significant characteristic changes from the time series and provide key information for subsequent crack risk assessment.

[0083] The specific implementation of step S60 is: according to the crack-prone area determined in step S20, the maximum change matrix ΔX obtained in step S50 is max Screening, get the attention matrix ΔX focus Specifically, first, ΔX max The absolute values ​​of the elements in are sorted, and the row and column coordinates of the first n% of the elements are selected as the coordinates of the focus area; then, according to the crack-prone area determined in step S20, the corresponding focus matrix ΔX is extracted. focus The purpose of this step is to associate the time series feature changes with the actual crack-prone areas and highlight the key regional features that have an important impact on crack risk. Usually n can be taken as 10% to 20%.

[0084] The specific implementation of step S70 is: the attention matrix ΔX obtained in step S60 is focusInput into the pre-trained crack prediction model to obtain a crack risk matrix, which is used to represent the crack risk score of each crack-prone area. The crack prediction model is trained using a supervised learning method, as follows: first, a batch of wall samples with cracks are collected, their multimodal image data is obtained and the fusion feature matrix is ​​extracted; then, according to the actual crack situation, the fusion feature matrix of each sample is labeled with a crack risk label, such as 0 for no cracks and 1 for cracks; finally, the crack prediction model is optimized and trained using labeled training samples using algorithms such as support vector machines (SVM) or deep neural networks (DNN). During prediction, the attention matrix of step S60 is input into the trained model to obtain the crack risk score of each crack-prone area. The purpose of this step is to establish a machine learning model that can accurately predict the risk of wall cracks.

[0085] The specific implementation of step S80 is: according to the crack risk matrix obtained in step S70, the crack-prone areas are classified, and the areas with higher scores are listed as key monitoring areas and output. Specifically, a crack risk threshold R is first set. th , the risk score is greater than R th The areas are divided into high-risk areas; then, the high-risk areas are sorted according to the risk scores, and the top m% areas are selected as key monitoring areas. The purpose of this step is to identify the areas that need to be focused on and monitored based on the results of the crack risk assessment, and provide a decision-making basis for subsequent repairs and preventive measures. th It can be around 0.7, and m can be around 20%.

[0086] In general, the above eight steps constitute a complete wall crack prediction process. Among them, steps S10-S20 determine the crack-prone area, steps S30-S60 extract and screen the key features reflecting cracks, step S70 trains the crack prediction model, and step S80 gives the crack risk assessment result. The entire method makes full use of multi-source heterogeneous data, and through the application of feature engineering and machine learning models, it realizes the accurate prediction of wall crack risks. Compared with traditional manual inspections or single sensor monitoring, this method has a higher degree of automation and prediction accuracy.

[0087] In order to better understand this method, a specific implementation process of this method is given below in combination with a specific formula:

[0088] The specific implementation of step S10 is as follows:

[0089] Based on the known wall structure information, such as wall material properties, geometric dimensions and support conditions, a three-dimensional finite element model of the wall is established. Specifically, the process includes the following steps:

[0090] 1. Geometric discretization

[0091] Discretize the three-dimensional geometric configuration of the wall and divide it into three-dimensional unit grids. For unit types, tetrahedral units and hexahedral units are commonly used. The shape function of the tetrahedral unit is:

[0092]

[0093] Where V is the unit volume, a i ,b i ,c i ,d i is the shape function coefficient. The shape function of the hexahedral element is:

[0094]

[0095] In the formula, are the coordinates in the local coordinate system of the element.

[0096] 2. Material parameter assignment

[0097] According to the characteristics of wall materials, each unit is assigned corresponding mechanical parameters, such as elastic modulus E, Poisson's ratio v, etc. These parameters reflect the mechanical behavior of wall materials.

[0098] 3. Determination of boundary conditions

[0099] Determine the boundary conditions of the finite element model based on the support conditions of the wall, such as simply supported, clamped, or hooked. The boundary conditions can be expressed by corresponding displacement, force, or constraint equations.

[0100] 4. Finite element solution

[0101] Using finite element solution methods, such as the stiffness method or the displacement method, physical quantities such as the stress distribution σ and the deformation field ε of the wall under various loads are calculated. The entire solution process can be expressed as:

[0102] [K]{u}={F}

[0103] Where [K] is the overall stiffness matrix, {u} is the displacement vector, and {F} is the external load vector. Through the corresponding composition relationship, σ = Dε, the stress field can be obtained.

[0104] Through the above steps, a three-dimensional finite element model that can reflect the true mechanical behavior of the wall can be established, providing a basis for the subsequent identification of stress concentration areas.

[0105] The specific implementation of step S20 is as follows:

[0106] According to the three-dimensional finite element model of the wall established in step S10, stress concentration areas and structural defect areas in the wall are identified through stress field analysis. The specific steps are as follows:

[0107] 1. Stress extraction

[0108] From the three-dimensional finite element model, extract the stress tensor σ of each unit:

[0109]

[0110] 2. Equivalent stress calculation

[0111] Calculate the equivalent stress σ for each element e , such as using the von Mises criterion:

[0112]

[0113] 3. Identification of stress concentration areas

[0114] According to the preset stress concentration threshold σ c , the equivalent stress σ e Greater than σ c The unit is divided into stress concentration areas. Usually σ c It can be taken as about 80% of the tensile strength of the wall material.

[0115] 4. Identification of structural defect areas

[0116] According to the known structural defect characteristics, such as holes, defect depth, etc., the units containing such defects are divided into structural defect areas.

[0117] 5. Determination of crack-prone areas

[0118] Combine the stress concentration areas and structural defect areas determined in steps 3 and 4 as the crack-prone areas in the wall.

[0119] Through this step, the crack-prone areas in the wall with stress concentration and structural defects can be identified, providing target areas for subsequent crack detection and prediction.

[0120] The specific implementation of step S30 is as follows:

[0121] At specified intervals (e.g., 1 hour), obtain and preprocess multi-mode image data of the wall, including ultrasonic reflection images, vibration transmission images, and surface infrared images. The specific steps are as follows:

[0122] 1. Ultrasonic reflection image acquisition

[0123] The pulse echo method is used to measure the internal structure information of the wall and obtain the ultrasonic reflection image. The echo signal s(t) can be expressed as:

[0124] s(t)=A(t)cos(2πf c t+φ(t))

[0125] Where A(t) is the amplitude, f c is the center frequency and φ(t) is the phase.

[0126] 2. Ultrasonic signal preprocessing

[0127] The acquired ultrasonic reflection signal is windowed using the time window function w(t) to filter out irrelevant interference:

[0128] s w (t) = s(t)w(t)

[0129] Commonly used time window functions include rectangular window, Hamming window, etc.

[0130] 3. Vibration transmission image acquisition

[0131] The vibration response of the wall surface is measured using a contact accelerometer, and the time domain signal is converted into frequency domain features using Fourier transform to obtain the vibration transmission image. The frequency spectrum of the vibration signal a(t) is:

[0132]

[0133] 4. Infrared thermal imaging image acquisition

[0134] The temperature distribution on the wall surface is photographed with an infrared thermal imager, and radiation correction and pseudo-color processing are performed to obtain a surface infrared image. The temperature T and radiation intensity I satisfy the Stefan-Boltzmann law:

[0135] I=εσT 4

[0136] Where ε is the emissivity and σ is the Stefan-Boltzmann constant.

[0137] 5. Multi-mode image data acquisition

[0138] Repeat steps 1-4 at specified time intervals (e.g., 1 hour) to acquire and preprocess a set of multi-mode image data including ultrasonic reflection, vibration transmission, and infrared thermal imaging.

[0139] The purpose of this step is to obtain multi-mode image data that reflects the internal structure, dynamic characteristics, and thermal characteristics of the wall, providing a basis for subsequent feature extraction and crack prediction.

[0140] The specific implementation of step S40 is as follows:

[0141] The pre-processed ultrasonic reflection image, vibration transmission image and surface infrared image are subjected to feature extraction and fusion to obtain a fusion feature matrix. The specific steps are as follows:

[0142] 1. Ultrasonic reflection image feature extraction

[0143] For the ultrasonic reflection image, the following features are extracted:

[0144] (1) Background noise characteristics n, available signal-to-noise ratio Characterization;

[0145] (2) Attenuation characteristic α, which can be modeled by exponential decay A(t) = A0e -αt Fitting;

[0146] (3) Echo peak feature A max , represents the maximum echo amplitude.

[0147] 2. Vibration transmission image feature extraction

[0148] For the vibration transmission image, the following features are extracted:

[0149] (1) Peak frequency f peak , corresponding to the main peak of the vibration spectrum;

[0150] (2) Power spectral density S(f), which represents the energy distribution of the vibration signal in the frequency domain.

[0151] 3. Feature extraction of infrared thermal imaging images

[0152] For infrared thermal imaging images, the following features are extracted:

[0153] (1) Temperature distribution mean

[0154] (2) Temperature distribution variance

[0155] (3) Temperature distribution gradient Indicates local temperature changes.

[0156] 4. Feature Fusion

[0157] The above multimodal image features are integrated into a fusion feature matrix X using methods such as principal component analysis (PCA) or independent component analysis (ICA). The objective function of PCA is:

[0158]

[0159] Where W is the principal component vector.

[0160] The purpose of this step is to extract features rich in crack information from the perspective of multiple physical quantities, providing a basis for subsequent crack prediction.

[0161] The specific implementation of step S50 is as follows:

[0162] For multiple fusion feature matrices obtained in time series, calculate the differences between matrices at adjacent time points to obtain a difference matrix sequence, and select the difference matrix with the largest absolute value as the maximum change matrix. The specific steps are as follows:

[0163] 1. Difference matrix calculation

[0164] Assume that the fusion feature matrix at a certain time t is X(t), then the difference matrix between adjacent times t and t+Δt is:

[0165] ΔX(t,t+Δt)=X(t+Δt)-X(t)

[0166] 2. Difference Matrix Sequence Construction

[0167] Calculate the difference matrices of all adjacent moments to form a difference matrix sequence {ΔX(t,t+Δt)}.

[0168] 3. Maximum change matrix identification

[0169] From the difference matrix sequence, select the difference matrix ΔX with the largest matrix norm (such as L2 norm) max , as the maximum change matrix. The matrix L2 norm is defined as:

[0170]

[0171] The purpose of this step is to extract the most significant characteristic changes from the time series and provide key information for subsequent crack risk assessment.

[0172] The specific implementation of step S60 is as follows:

[0173] According to the crack-prone area determined in step S20, the maximum change matrix ΔX obtained in step S50 is max Screening, get the attention matrix ΔX focus The specific steps are as follows:

[0174] 1. Maximum change matrix screening

[0175] ΔX max Sort the absolute values ​​of the elements in the image, and select the row and column coordinates of the first n% of the elements, which are the coordinates of the area of ​​interest. Usually n can be 10% to 20%.

[0176] 2. Focus on Matrix Extraction

[0177] According to the crack-prone area determined in step S20, the corresponding attention matrix ΔX is extracted. focus .

[0178] The purpose of this step is to associate the changes in time series characteristics with actual crack-prone areas and highlight the key regional characteristics that have a significant impact on crack risk.

[0179] The specific implementation of step S70 is as follows:

[0180] The attention matrix ΔX obtained in step S60 focus Input into the pre-trained crack prediction model to obtain the crack risk matrix, which is used to represent the crack risk score of each crack-prone area. The training process of the crack prediction model is as follows:

[0181] 1. Training sample preparation

[0182] A batch of wall samples with cracks are collected, their multimodal image data are obtained and the fusion feature matrix X is extracted. According to the actual crack situation, the fusion feature matrix of each sample is marked with a crack risk label y, such as y = 0 means no crack, y = 1 means crack.

[0183] 2. Model training

[0184] The supervised learning method is used to optimize and train the parameters of the crack prediction model using labeled training samples {X, y}. Algorithms such as support vector machine (SVM) or deep neural network (DNN) can be used. Taking SVM as an example, its objective function is:

[0185]

[0186] sty i (w T x i +b)≥1-ξ i ,ξ i ≥0

[0187] Where w is the weight vector, b is the bias, ξ i is the slack variable and C is the penalty parameter.

[0188] 3. Prediction Evaluation

[0189] During prediction, the attention matrix ΔX of step S60 is focus By inputting the trained model, the crack risk score r of each crack-prone area can be obtained to form the crack risk matrix R. For the SVM model, the crack risk score can be expressed as:

[0190] r=w T ΔX focus +b

[0191] For DNN models, the crack risk score can be the output of the last layer of the network.

[0192] The purpose of this step is to establish a machine learning model that can accurately predict the risk of wall cracks. Through the supervised learning method, using training samples containing crack information, the model parameters are optimized so that it can accurately transform the feature matrix ΔX focus Mapped to the crack risk score r.

[0193] The specific implementation of step S80 is as follows:

[0194] According to the crack risk matrix R obtained in step S70, the crack-prone areas are classified, and the areas with higher scores are listed as key monitoring areas and output. The specific steps are as follows:

[0195] 1. Risk threshold setting

[0196] Set a crack risk threshold R th , the risk score is greater than R th The area is classified as a high-risk area. th It can be around 0.7.

[0197] 2. Determination of key monitoring areas

[0198] The high-risk areas are sorted according to the risk scores, and the top m% of the areas are selected as key monitoring areas. Usually m can be around 20%.

[0199] 3. Output information

[0200] The information of key monitoring areas, such as regional coordinates, crack risk score r, etc., is output as a basis for decision-making.

[0201] The above calculation process uses the following variables: E: elastic modulus of wall material; ν: Poisson's ratio of wall material; σ: stress distribution in the wall; ε: deformation field in the wall; K: overall stiffness matrix; u: displacement vector; F: external load vector; D: material composition relationship matrix; σ: stress tensor; σ e : Equivalent stress of the unit; σ c : stress concentration threshold; s(t): ultrasonic echo signal; A(t): amplitude of echo signal; f c : center frequency of echo signal; φ(t): phase of echo signal; w(t): time window function; s w(t): echo signal after windowing; a(t): vibration response signal; A(f): spectrum of vibration signal; T: wall surface temperature; I: radiation intensity; ε: surface emissivity; σ: Stefan-Boltzmann constant; n: background noise characteristics; SNR: signal-to-noise ratio; α: attenuation characteristics; A0: initial echo amplitude; A max : Maximum echo amplitude; f peak : Peak frequency of the vibration spectrum; S(f): Power spectral density of the vibration signal; The mean of the temperature distribution; Variance of temperature distribution; Gradient of temperature distribution; X: fusion feature matrix; W: principal component vector of PCA; ΔX(t,t+Δt): feature difference matrix at adjacent moments; ΔX max : maximum change matrix; ‖ΔX‖2: L2 norm of the matrix; ΔX focus : attention matrix; n: percentage of attention area selection; X: feature matrix of training samples; y: crack risk label of training samples; w: weight vector of SVM; b: bias of SVM; ξ i : SVM slack variable; C: SVM penalty parameter; r: crack risk score; R: crack risk matrix; r th : Crack risk threshold; m: Selection percentage of key monitoring areas.

[0202] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for predicting cracks in a civil construction wall.

[0203] A third aspect of the present invention provides a civil construction wall crack prediction system, which includes the above-mentioned computer-readable storage medium.

[0204] Specifically, the principle of the present invention is:

[0205] 1. Establishment of three-dimensional finite element model: First, a three-dimensional finite element model is established based on known wall structure information, such as geometric dimensions, material properties, and boundary conditions. This model can accurately reflect the mechanical behavior of the wall and provide a basis for the subsequent identification of stress concentration areas. Through finite element analysis, key physical quantities such as stress distribution and deformation field of the wall under various loads can be obtained.

[0206] 2. Identification of crack-prone areas: Based on the stress field analysis results of the three-dimensional finite element model, the method of the present invention can identify stress concentration areas and structural defect areas in the wall. These areas are often the "soft ribs" where cracks occur and are the objects that need to be paid special attention to. By setting a reasonable stress concentration threshold, high stress areas and areas with defects can be divided into crack-prone areas.

[0207] 3. Multi-mode image data acquisition and preprocessing: In order to fully perceive the internal structure and mechanical state of the wall, the method of the present invention collects multi-mode image data including ultrasonic reflection, vibration response, and infrared thermal imaging. By preprocessing these data, such as filtering and transformation, various features reflecting crack information can be extracted from them, laying the foundation for subsequent feature fusion and crack prediction.

[0208] 4. Feature fusion and difference analysis: The method of the present invention uses feature fusion techniques such as principal component analysis or independent component analysis to integrate various features in multimodal image data into a comprehensive feature matrix. Then, by calculating the differences between feature matrices at adjacent time points, the most significant feature changes can be mined, providing key clues for crack risk assessment.

[0209] 5. Crack prediction model training: Based on the key feature changes obtained in step 4 and combined with the actual crack condition labels, the method of the present invention uses supervised learning methods, such as support vector machines or deep neural networks, to train a machine learning model that can accurately predict crack risks. This data-driven prediction method can better capture the complex laws of crack occurrence and development.

[0210] 6. Output of key monitoring areas: Finally, based on the output results of the crack prediction model, i.e., the crack risk scores of each crack-prone area, the method of the present invention classifies these areas, identifies the areas with higher risks as key monitoring objects, and outputs relevant information to provide a basis for subsequent maintenance decisions. By setting reasonable risk thresholds and selection percentages, we can focus on the most critical crack risk areas in a targeted manner.

[0211] In general, the key to the method of the present invention is to make full use of multi-source heterogeneous data, and achieve accurate prediction and monitoring of wall crack risks through technical means such as data fusion, feature extraction and machine learning modeling. Compared with traditional manual inspection and single sensor monitoring, this method has a better level of automation and prediction accuracy, and provides an innovative solution for effectively preventing and controlling wall crack problems.

[0212] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0213] In order to better understand the present invention, a specific embodiment is designed below. In the embodiment, the civil construction wall is made of reinforced concrete material, with dimensions of 12 meters long, 6 meters high, and 0.3 meters thick, and simple supported boundary conditions are adopted on all sides. During the use of the wall, local cracks have occurred, and effective monitoring and preventive measures need to be taken.

[0214] The crack situation of the wall will be analyzed in detail below according to the method steps proposed by the present invention.

[0215] Step S10: Establishing a three-dimensional finite element model

[0216] According to the geometric dimensions, material parameters and boundary conditions of the wall, the finite element method was used to establish the three-dimensional finite element model of the wall. Specifically, the geometric configuration of the wall was discretized into a tetrahedral unit grid, and the side length of each unit was about 0.1 meter. For the material parameters, the values ​​were taken according to the physical properties of reinforced concrete: elastic modulus E = 3.0×10 10 Pa, Poisson's ratio v = 0.2. The boundary conditions are set as simply supported conditions, that is, the displacement of the edges around is fully constrained.

[0217] After the three-dimensional finite element model was established, a static load analysis was performed on it, and the stress distribution cloud map of the wall under the action of its own weight was obtained. There are several stress concentration areas in the wall, mainly distributed near the support point and the middle position. These high stress areas are likely to be the weak links where wall cracks occur.

[0218] Step S20: Identify crack-prone areas

[0219] According to the stress cloud diagram obtained in step S10, the stress concentration characteristics in the wall are further analyzed. After testing, when the stress concentration threshold σ c Take it as 80% of the wall compressive strength, that is, σ c =1.6×10 7 Pa, the high stress area can be better divided out. These high stress areas are superimposed with the known structural defect areas (such as the holes of the embedded pipelines), and finally four crack-prone areas are determined, which are located in the four corners and the middle of the wall.

[0220] Step S30: Acquiring multi-mode image data

[0221] In order to fully perceive the actual conditions of the four crack-prone areas, ultrasonic reflection images, vibration transmission images and infrared thermal imaging images of the wall were collected. Specifically:

[0222] 1. Ultrasonic reflection image acquisition The pulse echo method was used, and a transducer with a center frequency of 50kHz was used to vertically measure the ultrasonic reflection signal inside the wall at 30cm above the surface of each crack-prone area. The measurement time interval was 1 hour, and the measurement was continuous for 24 hours, and a total of 24 ultrasonic reflection images were obtained. The original echo signal obtained was also windowed using the Hamming window function to filter out irrelevant interference.

[0223] 2. Collection of vibration transmission images A triaxial accelerometer was attached to the center of each of the four crack-prone areas to measure the dynamic response of the wall surface under environmental vibration. The sampling frequency was 1kHz, and the vibration signal was collected continuously for 24 hours. The time domain signal was converted into frequency domain features using fast Fourier transform (FFT) to obtain 24 vibration transmission images.

[0224] 3. Infrared thermal imaging image acquisition The infrared thermal imager was used to take thermal images of the surfaces of the four crack-prone areas under sunlight. The shooting interval was 1 hour, and the shooting was continuous for 24 hours, and 24 infrared thermal imaging images were obtained. For the original radiation brightness image, radiation correction and pseudo-color processing were also performed to highlight the temperature distribution characteristics.

[0225] Through the above steps, a total of 72 multi-mode image data (24 for each area) of 4 crack-prone areas were obtained within 24 hours. These rich physical quantity information laid a good foundation for subsequent feature extraction and crack prediction.

[0226] Step S40: Feature extraction and fusion

[0227] Based on the 72 multi-mode image data obtained in step S30, various features reflecting the state of the wall are further extracted and fused into a comprehensive feature matrix. Specifically:

[0228] 1. Ultrasonic reflection image features For each ultrasonic reflection image, the following three features are extracted: (1) Signal-to-noise ratio Among them A signal is the mean value of the echo signal amplitude, A noise is the mean value of the background noise amplitude. The signal-to-noise ratio reflects the clarity of the echo signal. (2) Attenuation coefficient α, using the exponential decay model A(t) = A0e -αt Fitting is performed, where A0 is the initial amplitude. The attenuation coefficient reflects the energy loss of ultrasound inside the wall. (3) Maximum echo amplitude A max , represents the peak amplitude of the echo signal.

[0229] 2. Vibration transmission image features For each vibration transmission image, the following two features are extracted: (1) Peak frequency f peak, corresponding to the main peak frequency in the vibration spectrum. (2) Power spectrum density S(f), represents the energy distribution of the vibration signal in the frequency domain.

[0230] 3. Infrared thermal imaging image features For each infrared thermal imaging image, the following three features are extracted: (1) Temperature distribution mean (2) Temperature distribution variance (3) Temperature gradient Indicates local temperature changes.

[0231] Finally, the principal component analysis (PCA) method is used to integrate the above eight features into a 72×8 fusion feature matrix X for subsequent difference analysis and crack prediction.

[0232] Step S50: Time series difference analysis

[0233] Based on the fusion feature matrix X obtained in step S40, the feature matrices at adjacent moments (one hour apart) were calculated to obtain 24 difference matrices ΔX(t, t+1). From these difference matrices, the one with the largest norm, ‖ΔX max ‖2=25.3, as the maximum change matrix.

[0234] This maximum change matrix ΔX max It reflects the most significant characteristic changes of the wall within 24 hours, providing important clues for subsequent crack risk assessment. max It was found that among the four crack-prone areas, the characteristics of the two areas near the middle (area 3 and area 4) changed most dramatically.

[0235] Step S60: Extracting the region of interest

[0236] According to the four crack-prone areas determined in step S20, the maximum change matrix ΔX obtained in step S50 max In the above example, the coordinates of the first 20% of the elements are selected as the coordinates of the focus area. focus , mainly concentrated in the two crack-prone areas in the middle (area 3 and area 4).

[0237] This result shows that although the four areas were identified as crack-prone areas in step S20, the time series analysis shows that the characteristics of the two middle areas have changed more dramatically, and there may be greater crack risks, which require special attention and monitoring.

[0238] Step S70: Crack prediction model training and evaluation

[0239] In order to accurately predict the crack risk of the identified areas of concern (areas 3 and 4), a batch of reinforced concrete wall samples with actual cracks were collected, and their multimodal image data and crack condition labels were obtained as training sets. Specifically, a total of 50 crack samples were collected, of which 30 were crack-free samples (label y = 0) and 20 were cracked samples (label y = 1).

[0240] The support vector machine (SVM) algorithm was used to optimize and train the parameters of the crack prediction model for this batch of training samples. The objective function of SVM is:

[0241]

[0242] Constraints: y i (w T x i +b)≥1-ξ i ,ξ i ≥0

[0243] Among them, w is the weight vector, b is the bias, ξ i is a slack variable, and C is a penalty parameter, which takes a value of 10.

[0244] After the training is completed, the attention matrix ΔX obtained in step S60 is focus The input was fed into the trained SVM model, and the crack risk score r of each area of ​​concern (area 3 and area 4) was output, as shown in Table 1.

[0245] Table 1 Crack risk assessment results for the concerned areas

[0246] area Crack risk score r Area 3 0.82 Region 4 0.75

[0247] As can be seen from Table 1, the crack risk scores of area 3 and area 4 in the middle are 0.82 and 0.75 respectively, both exceeding the preset threshold of 0.7, indicating that these two areas have a high risk of crack occurrence and need to be monitored and prevented.

[0248] Step S80: Output of key monitoring areas

[0249] According to the crack risk assessment result obtained in step S70, area 3 and area 4 with risk scores greater than 0.7 are determined as high-risk areas, and these two areas are sorted from high to low according to the risk scores, and the top 50% are selected as key monitoring areas.

[0250] Finally, the following key monitoring information was output:

[0251] 1. Key monitoring areas: Area 3, Area 4

[0252] 2. Crack risk score for region 3: 0.82

[0253] 3. Crack risk score for region 4: 0.75

[0254] This information can not only help to timely discover potential cracks in the wall, but also provide a scientific basis for subsequent maintenance and reinforcement.

Claims

1. A method for predicting cracks in a civil construction wall, characterized in that: The following steps are involved: S10, establishing a three-dimensional finite element model of the wall according to known wall structure information; S20, according to the three-dimensional finite element model of the wall, determining areas of stress concentration, structural defects, or cracked areas in the wall as key crack-prone areas; S30, acquiring and preprocessing multi-mode images of the wall at specified time intervals, including ultrasonic reflection images, vibration transmission images, and surface infrared images; S40, extracting and fusing the preprocessed ultrasonic reflection image, vibration transmission image, and surface infrared image to obtain a fusion feature matrix; S50, for multiple fusion feature matrices obtained in time series, calculate the difference between matrices at adjacent time points to obtain a difference matrix sequence, and select a difference matrix with the largest matrix norm from the difference matrix sequence as the maximum change matrix; S60, screening the maximum change matrix according to the crack-prone area to obtain a focus matrix; S70, inputting the attention matrix into a pre-trained crack prediction model to obtain a crack risk matrix for representing the crack risk score of each crack-prone area; S80, classifying the crack-prone areas according to the crack risk scores, listing the crack-prone areas with high scores as key monitoring areas, and outputting the results; Wherein, the step S60 specifically includes: Step 601: The maximum change matrix ΔX obtained in step S50 is max , sort by the absolute value of each element; Step 602, select the row and column coordinates of the first n% of elements, which are the coordinates of the focus area; Step 603: Extract the corresponding attention matrix ΔX according to the crack-prone area determined in step S20 focus .

2. A method for predicting cracks in a civil construction wall according to claim 1, characterized in that: The step S20 specifically includes: Step 201, extracting the stress tensor of each unit from the three-dimensional finite element model of the wall established in step S10; Step 202, calculating the equivalent stress of each unit; Step 203: According to a preset stress concentration threshold, the unit whose equivalent stress is greater than the stress concentration threshold is divided into a stress concentration area; Step 204, based on known structural defect features, including at least holes and defect depth, divide the unit containing such defects into structural defect areas; Step 205: Combine the stress concentration area and the structural defect area to determine the crack-prone area in the wall.

3. A method for predicting cracks in a civil construction wall according to claim 1, characterized in that: The step S30 specifically includes: Step 301, using a pulse echo method to obtain an ultrasonic reflection image of the internal structure of the wall; Step 302: Use a time window function to perform windowing processing on the ultrasonic echo signal to filter out irrelevant interference signals; Step 303: Use a contact acceleration sensor to measure the vibration response of the wall surface, and use Fourier transform to convert the time domain signal into frequency domain features to obtain a vibration transmission image; Step 304: Use an infrared thermal imager to capture the temperature distribution on the wall surface, and perform radiation correction and pseudo-color processing to obtain a surface infrared image; Step 305: Repeat steps 1-4 at specified time intervals to acquire and preprocess a set of multi-mode image data.

4. A method for predicting cracks in a civil construction wall according to claim 1, characterized in that: The step S40 specifically includes: Step 401, extracting background noise, attenuation, and echo peak features from the ultrasonic reflection image; Step 402: extracting peak frequency and power spectrum density characteristics of the vibration spectrum from the vibration transmission image; Step 403: extracting the mean, variance, and gradient features of the temperature distribution from the infrared thermal imaging image; Step 404: Use principal component analysis or independent component analysis to integrate the above multi-modal image features into a fusion feature matrix.

5. A method for predicting cracks in a civil construction wall according to claim 1, characterized in that: The step S50 specifically includes: Step 501: Let the fusion feature matrix at a certain time t be X(t), and calculate the difference matrix ΔX(t,t+Δt)=X(t+Δt)-X(t) between adjacent times t and t+Δt; Step 502, calculate the difference matrix of all adjacent moments to form a difference matrix sequence {ΔX(t,t+Δt)}; Step 503: Select the difference matrix ΔX with the largest matrix norm from the difference matrix sequence max Maximum change matrix.

6. A method for predicting cracks in a civil construction wall according to claim 5, characterized in that: The step S70 specifically includes: Step 701: The attention matrix ΔX obtained in step S60 focus Input into the pre-trained crack prediction model; Step 702: The crack prediction model outputs the crack risk score of each crack-prone area to form a crack risk matrix.

7. A method for predicting cracks in a civil construction wall according to claim 6, characterized in that: The step S80 specifically includes: Step 801: Set a crack risk threshold R th , the risk score is greater than R th areas are classified as high-risk areas; Step 802: sort the high-risk areas according to their risk scores, and select the top m% of areas as key monitoring areas; Step 803: Output the information of the key monitoring area.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for predicting cracks in a wall during civil construction as described in any one of claims 1 to 7.

9. A civil construction wall crack prediction system, characterized in that: Contains the computer-readable storage medium of claim 8.

Images